The effects of interference on monopulse performance...VI, V2/V1 and F3z - f?l from (6). Before per...
Transcript of The effects of interference on monopulse performance...VI, V2/V1 and F3z - f?l from (6). Before per...
MASSACHUSETTS INSTITUTE OF TECHNOLOGY LINCOLN LABORATORY
THE EFFECTS OF INTERFERENCE ON MONOPULSE PERFORMANCE
R. J. McAULAY
TECHNICAL NOTE 1973-30
1 AUG 1973
LEXINGTON MASSACHUSETTS
.
CONTENTS
List of Illustrations
1. Introduction
2. Interference Effects on Estimate Bias
3. Special Case: Real Antenna Patterns
4. Interference Effects on Estimate Variance
5. Conclusions
References
Appendix
Acknowledgment
. .
~.,
,..111
iv
1
2
8
21..
27
29
30
31
LIST OF ILLUSTRATIONS
Figure
l(a) EQUIVALENT AZIMUTH USINGTHE MONOPUI.,SE FUNCTION. . . . . . . . . . . . . . . . . . . . . . . ...12
l(b) PROBABILITY Y DISTRIBUTIONFUNCTION OF EQUIVALENT AZIMUTH. . . . . . . . . . . . . . ...12
2. AZIMUTH ESTIMATE IN THE PRESENCE
OF INTERFERENCE. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3. AZIMUTH ESTIMATE IN THE PRESENCE
OF INTERFERENCE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4. VARIANCE OF THE AZIMUTH ESTIMATE
IN THE PRESENCE OF INTERFERENCE . . . . . . . . . . . . . . . . 26
iv
,,
1. INTRODUCTION
There are several competing factors that limit the performance of the
monopulse azimuth estimation accuracy in a noise and interference background.
The first is due to the fading phenomenon that results in a degradation in the
signal-to -noise ratio (SNR) as the target and interference signals approach
the out of phase condition. Pruslin [1], in a simulation of the receiver noise
effects on monopulse performance, has observed that the estimator becomes
biased for low values of the SNR. Browne [2], in an interference-free analysis,
has calculated this bias exactly. for all values of the SNR.
Closely related to the bias introduced by fading, is what we shall refer
to as the azimuthal bias that results when the interferer and the target signal
sources are located at different azimuths. Finally, the presence of inter -
fering signals causes the variance of the estimate to increase.
In this note we shall study mainbeam and sidelobe interference and
noise effects simultaneously and use an analysis similar to Browne’s to
compute an exact expression for the estimator bias. We specialize our results
to real antenna patterns and obtain an expression for the bias that holds for
the high and low SNR situations, hence, we obtain the azimuthal bias and the
fading bias simultaneously. Numerical results are given for some typical
operating conditions and it is shown that the bias can be a large fraction Of
a beamwidth.
Some further analytical results are obtained for the case of sidelobe
interference for real antenna patterns. In this case too, it is possible for an
ATCRBS (Air Traffic Control Radar Beacon System) interferer that is located
in a large near-in sidelobe to cause bias errors that are a large fraction of
a beamwidth.
1
In addition to introducing a deterministic bias to the azimuth estimate,
interference causes the random error to have an increased variance.
We obtain a general expression for the variance that applies to the mainbeam
and sidelobe interference cases using an analysis similar to that developed
by Sharenson [3] who analyzed the noise-only case. Numerical results arc
given that show that the variance also depends critically on the relative
phase between the target and interference signals.
These general expressions for the bias and the variance of the mono-
pulse estimate provide the tools needed to thoroughly understand the effects
of interference. These are essential in the evaluation of signal processing
techniques. to overcome the effects of interference. The leading candidate
at the moment, is the monopulse interference detector and data editing
s theme followed by an outlier te st. Although this is the subject of a sepa-
rate paper [4] the results of the present study are the background for the
succeeding analysis.
2. INTERFERENCE EFFECTS ON ESTIMATE BIAS
We restrict our attention to sum-difference (even-odd) monopulse pro-
cessing that is performed on a sampled-data basis. Assuming mixer pre-
amplifiers at the output of the sum and difference beams, the received si,g -
nals in the presence of interference are modelled by
‘.,
jvB j VIyi = A=e Gi(e~) + Ale Gi(E!l) + n. i=l,2
1(1)
where yl, y2 refer to the outputs of the sum and difference beams; As, q~,
es and AI, t+iI, f)l are the amplitude, phase, and azimuth of the target and interference
respectively; Gl( ), G2( ) are the antenna patterns of the sum and difference beams,
and these may be complex in general; nl, n2 are the independent, zero mean Gaussian
noise samples due to the mixer preamps whose real and imaginary parts have vari -
2ante o
An azimuth estimator that is often used in practice is given by
the relation
E(j) = Re(y2/yl) (2)
where the monopulse function, E ( e), is simply the normalized difference
pattern,
E(e) = G2(9)/Gl(9) (3)
In most cases of practical interest, this function is well approximated by
a linear characteristic, hence we can write
E(13) = k(6/BB) (4)
where 9B
refers to the 3 dB beamwidth of the antenna patterns and k is a
standard parameter that arises in the characterization of monopulse systems.
Usually 1 S k S 2 with k . 1.5 being a typical value. When this approximation
is valid, the azimuth estimate is readily generated as
(5)
We will incorporate (3B into our definition of e and hence express all our
results in units of a 3 dB beamwidth.
Our goal is to compute the mean and variance of (5) when inter-
ference is present in the sum and difference signals as given by (l). We
begin by letting
j US j VI .Di
Ui = A~Gi(9~)e + AIGi(fI1)C = Vi# (6)
Since ni are complex Gaussian variates, they can be expressed as
ni . Ni expj a; where Ni are Rayleigh and O; are uniform random variables.
Using these definitions in (1) we have
(7)
4
Since ~i and W> are independent and uniform, then so too are o ;-pl~ ~,,
We apply these relations to evaluate (5) by writing
Re (L\. ‘e(Y~Y~)
.
\yl) IYJZ
1( )[ 1V2COS(B2 - 81) + NZc05~2 VI + NlCOSCI1 + ‘~sin(~2- Bl)+N2sin% ‘lsin~
.L L
‘1 + 2vlNlc0s% + ‘1 (8)
Following Brov.me [2] we first average (8) over IY2. Since this is uniform on
(O, 2n), we have
E (k$) =V2COS(~ - 81) (Vl + NICOE~) + V2N1sin(~2 - ~l)sintrl
%
(9)
V; + N: + 2T1N1COSCY1
Next he averages over O’1. In the Appendix this is shown to give
E (k~) =
al ‘2
o if N1>V1
‘2(lo)
—cOs (@2 - PI) if N1<V1
‘1
5
Since N1 is the magnitude of a complex Gaussian random variable having
2.variance 20 , It has the Rayleigh distribution
N;2
()
‘1p(N1) = — exp - ‘2
tiu 40
Then averaging (10) over N1 gives
‘2E(kil) = ~ COS(D2-5 ) A
1 1 fill
v. [
(11)
v~1
()
- N:
N1 exp7
dN
(1 401
(12)
It should be noted that when interference is absent, AI = O
and from (6) we see that ~1 . ~ and (12) reduces to the same expression
obtained by Brovme. However, we are now in a position to determine the
effects of interference on the azimuth bias. To do this we need to evaluate
VI, V2/V1 and F3z - f?l from (6).
Before per formi~ this evaluation wc first note that the target of
interest lies within the mainbeam of the antenna, hence we may assume that
G1(fl~) and G2(6~) are real functions of 9 Whereas the sum and differences“
beams are in phase over the antenna beamwidth, small errors in the ampli-
tude and phase tapers render them complex in the sidelobes. Since the
6
interference can be located in the mainbcam or within the sidelobes of the
antenna, then Gi(61) must be considered to be complex in general. To make
this explicit we write
Gi(~) = Ai ($) expyi (91)
Then substituting this into (6) we can easily show that
1Vi = A:G~ (o.) + 2A~A1Gi (o.) Ai (91) ~OS [v+ Yi(el)]
(13)
(14)
-1 AIAi ($) sinw+ yi (eI)ei = ,0~ + tan
A~Gi(8~) + A#i(8~cos [CO+ Yi($)](15)
where v = q - co. is the relative phase between the tar~et and intcrfercncc
signals. Using these expressions in (12) and performing straightforward
but tedious trigonometric manipulations we can show that the first moment
of the azimuth estimate is
(16)
where p . A1/A~ represents the intcrfcrcn cc-to-signal ratio (ISR) prior
to any processing and
7
where
-1y . tan
Az
2 ‘1l+ZO+COS(Q+Y1)+P y
1‘1
G2(B~) Al sinl’l +G1(Eis) A2 sill y2
G2 (es) ‘1 Cos ‘1+- G1(e~) A2cos y2 1
\
(17)
(18)
These are the key expressions we need to fully understand the deterministic
aspect of the azimuth error introduced by interference.
3. SPECIAL CASE: REAL ANTENNA PATTERNS
The results that have been obtained to this point are free of any
approximating assumptions and apply to the general case of complex antenna
patterns. Unfortunately it is difficult to draw analytical conclusions from
these expressions because of the large number of parameters that would have
to be examined simultaneously. Results for the general case will be obtained
using a computer simulation. There is one case of considerable practical
8
I
I
.
importance that we can explore analytically. This is the situation in which the
antenna patterns are real, which is a reasonably accurate model for the high
near-in sidelobes of the antenna. For this case wc assume that
Ai(91) si,, Ti(131) % O i=l,2 (19)
and then define
Gi (131) = Ai (fII) CCJSYi ( 91) (20)
which we note arc real functions for all values of tll. Using this assumption,
and (17) become
iie.i0
[[
+;(0s) Gl(f+) ~ G;(eI)l-exp- 1 + 2Pmc0scp+ p
2.2 G; (es) IIwhere now
0G1(131) ~ G:(B1)
1+2PG1(9~) co’ ‘“+ pG:(9~)
(21)
(22)
9
We recall that p = A1/A~ represents the interference-to-signal ratio (ISR) prior
to any processing. The measured ISR at the outputs of the antenna ports is
given by
A1G1(91) _ G1(51)P. ‘
A~G1(8~) - p G1 (9~)(23)
Although both definitions of ISR are important parameters of the system, it
is convenient to rewrite (21) and (22) in terms of p . In this case the mean0
value of the azimuth estimate is given by
j= $ 1[ -A:G; (13~)1 - exp
0II
(1 + Zpocosw+ p:) (24)
2 U2
where now
G2(8~)
[
G2(13~)G2(eI)
1
z G2(e1)
—+p ——G1(8~)
cos~ip —o G1(13&) + G1( 61) o G1(el)
kg =0
1 + 2poc0sm+ p:
It is interesting that the performance depends
E(fl) = G2(0/G1(13). In most cases of interest
(25)
only on the monopulse function
the target will be located within
the 3 dB beamwidth of the antenna. Therefore, using (3) and (4) we can write
G2(8~)
Gl(e~) = ‘es(26)
.
10
where 9~ is measured in 3 dB beamwidths. In the case of mainbeam inter-
ference this linearity property will also be valid, but in more general cases
E(81) will simply take on the values of the normalized difference pattern. 10
handle this situation we define an equivalent az.i]muth
Gz(O1)
F1=~—k Gl(t31)
(27)
and note that whenever el is within the 3 cll> beamwidth of the antenna 9 = 6I 1“
In Figure l(a) we have plotted the equivalent azimuth for a monopulse antenna
configuration that has a 4° beamwidth and achieves a -20 d13 peak sidelobe level
on both the “sum” and “difference” beams. Assuming that ~1 is uniformly
distributed in (-TI, n), Figure l(b) shows the probability distribution function
of the equivalent azimuth. This shows that “most of the time” the equivalent
azimuth will be less than 2 beamwidths, hence, for the purpose of analysis,
wc can study the cases of mainbeam and sidelobe interference simultaneously.
Then using (26) and (27) in (25) we can express the first moment of the azimuth
estimate as
xxe=e I1- exp
o
A:G:(fJ~)
(l+2poc0sw+p
202~2) II
x e~+Do(e~+F1)cos(, +p:F1e. =
1+200 CO SW+O:
(28)
(29)
11
20
0
-2C
4° BEAM WIDTH
-20-dB PEAK SIDE LOBES
(o)
1 I I I
--@EEL
I I I
-80 -60 -40 -20 0 20 40 60 80
AZIMUTH (deql
Fig. la. Equivalent azimuth using the monopulse function.
!.0 r 1
J’”
-6 -4 -2 0 2 4 6
EQUIVALENT AZIMUTH ( Beomwid!hs)
Fig. lb. Probability distribution function of equivalent azimuth.
12
It is from this equation that we can begin to make some general
statements regarding the performance of the azimuth estimator. We shall
first show how the results arc consistent with th~se previously obtained and
then go on to consider cases that have not yet been explored in the literature.
(a) Case 1: No Interference
When there is no interference, AI . 0, hence P = O and fromo
(28) and (29) the average value of the estimate is
[ IIA;G:(3~)
l-cxp -
2D2(30)
This is the result obtained by Browne [2] and describes analytically Pruslin’s
observation [1] that as the SNR, A sz G ,2 (9 s)/2 o 2,dccreascs, the monopulsc
estimate is biased towards the antenna bores ight.
We have tabulated (30) in Table 1 from which it can be seen that
for SNR’S above 8 ~B, the bias can be considered to bc negligible. Since in
the Air Traffic Control (ATC) system the SNR will be at least 20 dB, this
effect will be insignificant unless fading occurs, as willbe discussed in the
next section.
Table 1. Variation of Bias with SNR
SNR (dB)
o3
3.6
4.766.6
7.258.39
8.89.64
iie-e
s
eB
0.370.135
0.1
0.05
0.01
0.005
0.001
0.0005
0.0001
(b) Case 2: Fading
If the target and interference signals arrive from the same azimuth
(multipath from a flat earth for example) then El ❑ $ = 8~ and (28) and (29)
reduce to
xe=e
s[
A: G; (85)l-exp- (1 + 2p&0smi P:)
2’02 II
(31 )
This demonstrates the existence of a fading bias that can occur even at large
SNR when the interference
tends to cancel its energy.
fading will be negligible as
This will be the case if
signal occurs out of phase with the target and
From Table 1 wc conclude that the bias due to
long as the effective SNR is greater than 8 d13.
14
For a 20 dB
.752 p. ~
(c) Case 3:
20 ‘
SNR, this shows
1.25.
Azimuthal Bias
+ Zpocosm + & ~ 6.3 (3~,
that fading could become a problem only if
. .Next, we consider the case in which the SNR is large and the
interference is of low enough power that the exponential term in (28) is
negligible. Then (29) reduces to
x es + (es +F1)oocOsm + p. ~‘F
e= (33)
1 + Zpocosw+oz0
This result is sufficiently general to describe the cases of mainbcam and
xsidelobe interference. Notice that when OS = 81 = ~1 then 8 ‘ es and the
estimator is unbiased. T’hercfore a bias is obtained only when f3~ # e ~ ,
namely when the target and interference signals are separated in azimuth.
It is for this reason that wc refer to this effect as the azimuthal bias.
To obtain some physical appreciation for the behaviour of the
estimator wc consider a specific case of interest. We assume the
target is located on boresight at 20 dB SNR. Therefore, es = O and
A~G; (0)/2u2 = 100. Wc next let the interferer be located at the edge of the
3 dB beamwidth so that &l = ff, = .5. TO within a good approximation the sum
beam is quadratic over the bcamwidth, hence G1(9) = 1 - 1.7132. Since the
15
output interference-to-signal ratio is PO = pG1(O1)/G1(e~) where p = A1/A~,
wc see that p = . 707p Equation (33) is plotted as a function of the relative0
phase angle for various values of p. The results are shown in Figure 2.
Equation (28) was also plotted but there was no discernible difference in
resulting curves at least for the values of p we used. In Figure 3 these
the
equations were again plotted for P. nearer to unity and the fading effect can
be observed at the out of phase condition. Since there is only a small range
of values for which this effect is observable and since its effect is, if anything,
beneficial, we shall neglect the fading bias in the rest of our work and restrict
our attention to the azimuthal error as described by (29).
The bias in the estimate can be written as
—P. + Cos m
b(v) : : - El, = (61 - es) P.
1 + 2POCOSO+P20
where PO and 91 are given by (23) and (27) respectively.
(34)
Therefore for small
t. 0x
values of p. , ~hilc for Iargc values B . ~1 which shows how the inter-s’
ferer “captures” the estimate. For intermediate values, curves like those
shown in Figures 2 and 3 are obtained which demonstrate the so-called scin-
tillation effect which is a term used to describe the fact that the azimuth
estimate lies outside the azimuth interval ( Bs, f31).
16
-IEzm
SIR (C
ATCRBS AZIMUTH
1
ADABS AZIMUTH
SNR = 20dB0,=0
81 = 0,5Bw
I
0 Tl Z7T
-m
-40
-3
3
to
.+m
RELATIVE PHASE BETwEEN TARGET AND INTERFERENCE SIGNALS (rod)
Fig. 2. Azimuth estimate in the presence of
17
interference.
4
v,—3E
:m
Lomu0
2
0
-2
.,
lEzEuIl,.NO FADING
FAOING
h /
I \
I I
SIR= -fdB
/:~:::~H
\ DABS AZIMUTHSNR= 20d B
0,=0.5BwSIR =ld B
\1“/
&. .-“ ,, c ,,
RELATIVE Pti ASF BETWEEN TARGET AND INTERFERENCE SIG NALS (rod)
Fig. 3. Azimuth estimate in the presence of interference.
18
(d) Case 4: Effects of Relative Phase
It is reasonable to model the phase as a uniformly distributed
random variable on (O, 2TT). Then averaging over phase, it is easy to show
,. that
J_
2TI
‘h
b(v) d~ =
o ifpo<l
1 ifp>l0
(35)
In a practical situation, the key issue is how correlated the phase is from
sample to sample within a reply. For multipath, the correlation time can
be several seconds because the phase relationship depends on the relative
path lengths between the direct and multipath signals and for typical aircraft
speeds, these change slowly. For ATCRBS interference, the phase difference
will be independent from sample to sample since lhe transmitter tube is in-
coherent from pulse to pulse. IIowcver, since there arc relatively few inter-
ference samples in any one reply, the inherent averaging due to phase
cannot reliably be exploited and wc must face the possibility of having to
deal withthe large errors shown in Figure 2.
19
(e) Case 5: Sidelobe Intcrfcrcnce
Although all of the results given so far are applicable to mainbcam
and sidelobe inter fcrcncc, there arc some remarks that are worth noting for
the latter case. In the case of A’1’CR}~S sidelobe inter fercncc, it is possible
that A1/A~ can be much larger than unity. I,ct us SUpPOS~ fOr example that the
DABS target is at maximum range (100 mi) and the A1’CRl\S interferer is at
the same power, but at a C.1OSC in range (10 mi say). Then the 20 dl’J decrease
in ISR due to the antenna sidelobes is compensated by the 20 cl]> increase in
power due to the range diffcrcnc. c. I’hcn Do can be near O dR and since the
equivalent azimuth is of the orclcr of 1 or 2 beamwidths then, as we have
seen, large bias errors can be cxpcctcct. Therefore wc conclude that if a
strong AT CRBS signal is being reccivcd in a near-in siclclobc, the monopulsc
azimuth estimate can have a lar~c bias that exhibits the same clependcncc on
the relative phase bctwccn the DABS and AT’CR13S transponders that was
shownto exist for mainbcam interference. Furthermore, it is clear that very
low sidelobcs is not a sufficient means of eliminating the effects of this inter-
ference. Therefore, at least one additional lCVC1 of data processing will be
needed to improve the quality of the azimuth estimate for tbcse cases. T’his
is referred to as monopulsc data editing and will be discussed in a later
paper [4].
20
I .
For sidelobe multipath, on the other hancl, wc can reasonably expect
that the reflection coefficient, A1/A - .707. For antenna patterns with 20 dBs
peak sidelobe levels, this puts PO less than . 07. From our studies so far we
know that the bias error peaks when the direct and multipath signals are w out
of phase. Using (34) the bias error can be bounded by
(36)
where the last approximation follows from the fact that o is quite a bit lCSSo
than unity, For a DABS target on horcsight, 86 = O and then using (23) and (27)
the bound be comes
(37)
which shows that the bias error depends on the sidclobe level Of the differ-
ence pattern relative to the sum beam gain, For the example studied, k = 1. 5,
A1/A~ .. 707 and G2( 91)/ Gl(~~) = . 1, which yields a peak bias error .047
3 dB beamwidt.hs.
4, INTERFERENCE EFFECTS ON ESTIMATE VARIANCE
In this section we will compute the variance in the monopulse azimuth
estimate. Sharenson [3] has studied this problem for the case when the inter-
ference consisted only of receiver noise. We shall extend his analysis to in-
clude the effects of mainbeam and sidelobe interference. We begin with tbe
equivalent signal model formulated in (l), (2), and (5). The monopulse estim-
ator of interest is
21
(38)
which gives the azimuth estimate in 3 dl> beamwidths. l“rom (1) and (6),
so that
yi = lJi + n.1
lY1f= lu~
where the last approximation holds provided t.hc equivalent SNR, IU1 [ 2/202
is large enough, typically >12 dB. Then the moments of (38) can be com-
puted by evaluating the moments of
>k
7, = Rc(ylyz ) (41 )
Since the noise terms nl and nz are inclcpcndcmt, the first moment is simply
.,.
( ‘)z ‘ “ul~z(42)
and hence the first moment of the azimuth estimate is given approximately by
K ‘2e=&c0s(B2-P1)
1
(43)
22
. .
which is the same result wc obtained in the last section when the effective
SNR was greater than 8 dB. Proceeding further, the second moment of (41)
>~can be calculated using the identity Re(x)R@ = $ Re(x Y ) + ~Re(x Y.) In this
>:case we let x= y. ylyz and then
— ——
2z= ~ly1121y2]2+~Rey~y~2
2 TFrom (39) and using the fact that Inil = 202, ni = O, wc obtain
lYi12 = lui12 + 204
72Yi ,= LJ.
Combining (42), (44) and (45 ) wc can show that the variance of z is
~ -2-z
(= IIJ
(44 )
(45a)
(45b)
)12+. \u212 02+204 (46)
Finally we use (4o), (M ) and (M ) tO shOw that the variance Of the azimuth
estimate is
,-[14.-+.-]Var (6) = ‘z (47)
23
Since [ui I = Vi, then (47) could bc expressed in terms of the complex antenna
pattern parameters by using (14). Unfortunately the resulting expressions
are complicated and in order to obtain sornc ~meaningful analytical statements
it is not fruitful to retain the most general equations. Therefore, we shall
.,follow the direction of the proceeding section and restrict our attention to the
case of real antenna patterns. In addition wc note that (47) describes the‘,
estimate variance only when the cffectivc SNR is large, This means that
lu112/202>x, hence wc can also neglect the effects of the second order
SNR term that appears in (47). Under these conditions we can then use (14)
for real antenna patterns and show that the variance reduces to
1 +k2’Ff~i$l~- 2
z l+k61
22l+2p cos~ig
2 l+k 9 0 l+k2’r3j0
1 + k2Ei2Var (~). z: .
s s(48)
A~G1(9~) k22
l+2poc0sq+po
where, as in the case of the bias effects, O. ‘“ arc given by (23) and (27).and 01
In the next few paragraphs we shall consider some more specialized
,,cases.
(a) Case 1: No Intcrfcrencc.,
When there is no intcrfercncc, AI . 0, hcncc p . 0 and from (48)0
we sce that the variance is simply
24
2 1 + k28;a
Var ($) = .
A~G;(8~) k2(49)
which is the result obtained by Sharenson [3] and others, and shows that the
variance decreases with SNR and the slope of the normalized difference
pattern and increases with the degree to which the target is off bores ight.. .
(b) Case 2: Fading
If the target and interference arrive from the same azimuth
(multipath from a flat earth for example) then %1 = 91 = eq and (48) becomes
2 l+k282
Var ($) =a s 1
(50)
A~G; (96) k2 1 + 2 PO COSO+. P20
which shows that when multipath fades occur there can be a reduction in the
effective SNR which in turn leads to an incrcasc in the variance. In the last
section we found that a bias error was also introduced in the fading situation.
(c) Case 3: Azimuthal Variance
When the target and interferer are at different azimuths we see
from (46) that the variance will bc further increased. Typical results arc
.,plotted in Figure 4 for the case of a target on boresight at 20 dB SNR
and an interferer located at the edge of the 3 dB beamwidth. We see that the
25
0EOa0
SNR=20d B
Q*=O
t?l =0.5Bw fl SIR (dB)
(
Za 0 0.577 n 1.5T 2T~
RELATIvE PHASE BETWEEN TARGET ANO INTERFERENCE SIG NALS(rad)
Fig. 4. Variance of the azimuth estimate in the presence of interference.
most significant degradation in performance occurs at the out-of-phase con-
dition. However, the magnitude of the errors is much smaller than the con-
tribution due to the bias, and furthermore since the errors are random,
their effect can bc further reduced simply by averaging many
,. Therefore, we conclude that the bias error is likely to be the
,. problem for ATC direction finding.
of the estimates.
more troublesome
5. CONCLUSIONS
An exact expression for the bias of a two beam monopulse azimuth
estimate has been derived that describes the degradation in performance due
to receiver noise and AT’ CRBS and multipath intcrfcrencc. At low SNR and
high ISR the bias tends toward borcsight, although for all practical purposes
the net effect of this bias is negligible. More important is the azimuthal bias
that describes tbe so-called scintillation of the azimuth estimate that has
been observed in many monopulsc tracking problems. Depending on the
values of the SNR, ISR and the relative phase between the DA}IS and inter-
ference signals, the bias can bc quite significant even when the inter fcri.ng
signal arrives through a low lCVC1 sidclobe,
A first order analysis was used to obtain the variance in the.
azimuth estimate when intcrfcrcmcc is present in a noisy background.
Although the results indicate that the random errors will not be insignifi-
cant, the deterministic. bias error will bc, by far, the more dominant effect.
For ISR’S less than -30 cl}~, the effect of the interference is negligible.
As this quantity increases, the bias and variance increase and errors that are
of the order of a beamwidth can bc obtained. For ISR’ s between + 10 dIl there—
is a strong dependence on the relative phase, with a peak error occurring at
the out-of-phase condition. As the ISR is further inc. rcased, this peaking
effect subsides until the intcrfcrcncc completely captures the monopulse>,
processor which at this point WOUIC1track the intcrfcrcncc target..,
Although the results arc applicable to analyzing the effects of
A1’CRBS interference and multipath, a distinction between the two phenomena
should be noted. For mu]tipath, the direct ancl indirect signals are at the
same frequency and coherent in the sense that their relative phase may be
constant during several seconds duration. For A’1’CRIH interference,
however, the transponders arc incoherent from bit to bit and possibly for
samples within a bit since there may bc carrier frequency offset. ‘1’hercforc,
in the AT CRBS case, there will bc averaging that can be exploited to reduce
the overall bias error, in which case the effect of the increased error
variance would become a more important effect. From a processing point
of view, this could be ovcrcomc by using more samples to form the azimuth
estimate.
In a separate study [4] the pcrformanc. c of the maximum likelihood .
interference detector has been prescmtcd in detail. T’he next step is to com -
bine the results of both studies to evaluate the data editing concept. T’his
will determine whether there is any promise to the idea of introducing an
additional level of data processing to improve the overall quality of the
azimuth c stimatc.
28
1. 1>. 11. Pruslinr l,M”ltipath.Ikllse Monopulse Accuracy, “
Lincoln l.,aboratory “1’ethnical Report 487, (I9 November1971). DDC AI>-737369.
2. A.A. L. l\rownc, l,incoln J,aboratory Internal, Memorandum.
3. S. Sharenson, “Angle Wtimination Accuracy with a MonopulseRadar inthc Search Mode, ,HIRE Transactions on Aerospace
and Navigational Electronics (September 1962), pp. 175-179.
4. R. J. McAulay and 1’. P. McGarty, “Maximum LikelihoodDetection of Unresolved Radar Targets ancl Multipath, “ tobe published.
5. IJ, K. Barton, “Raclar System Analysis, “I’rentice - Ilall, NewJersey, 1964.
. .
APPENDIX
To compute the average of (9) with respect to al, a uniformly
distributed random variable on (O, 2TT)we have
2Tl 2
E (k~) = C0S(~2- ~1)
.r
‘l+vl NlcOswl
% ‘2d al
20 V;+N1+2V1N1 .Os O1 I
Since
zero.
the integrand of the second term is an odd function, its integral is
Using standard integral tables, 13rowne showed that the integral in the
first term reduces to
V; -N:2TI
‘2.os(~-B1) -&. ~
f
1da
( )1
V;+N; o ~+2V1N1 1
— Cosci
V;+N:1
1 ‘2
~[
V; -N;—.l-— COS(B2-B1) 1+ V2+N2
7 VI
11
30
~v2N2
II1/2
11
(V:+ N;)2
.
where the positive square root is under stood.. Therefore
if N1<V1
as given in. (10).
ACKNOWLEDGMENT
The author would like to acknowledge the unpublished work of A. A. 1..
13rowne of the Lincoln Laboratory who first derived the exact expression for
the first moment of the monopulse estimator in a receiver noise background.
The author would also like to thank D. F. DeLong who provided Brownc’s
obscure but important memo and whose comments, along with those of
E. Kelly, resulted in a considerable improvement in the final draft. Final~y.
the author acknowledges the related but unpublished work done in this area
by J. Modestino.
31